Jensen Huang stood in front of 18,000 people at San Jose’s SAP Center on Monday, wearing his signature black leather jacket, and casually dropped a number that would make most Fortune 500 CEOs choke on their coffee: $1 trillion.

That’s the revenue opportunity Nvidia now sees for its AI chips through 2027 — doubled from the $500 billion estimate it gave investors just last month. And after a nearly three-hour keynote that covered everything from space-based data centers to Disney robots to the future of gaming graphics, one thing is crystal clear: Nvidia isn’t just riding the AI wave anymore. It’s building the ocean.

Here’s what actually matters from GTC 2026.

The Inference Inflection Point Has Arrived

The big strategic shift that got buried under all the hardware announcements: Nvidia is pivoting hard toward inference — the process of actually running AI models — and away from its traditional dominance in training.

“The inference inflection has arrived,” Huang declared from stage.

Why does this matter? For years, the AI gold rush was about building models. Companies like OpenAI, Anthropic, and Meta spent hundreds of billions training their AI systems. But now that those models exist, the money is shifting to deploying them at scale — serving hundreds of millions of users hitting ChatGPT, Claude, and Meta AI every single day.

Training is a periodic cost. Inference is an ongoing, ever-growing expense that scales with every new user, every new query, every AI agent running in the background. And Nvidia wants to own that market too.

The problem? Inference isn’t as GPU-dependent as training. Traditional CPUs from Intel and custom chips from Google are legitimate competitors. Which is exactly why Nvidia’s biggest announcement wasn’t another graphics card — it was a CPU.

Vera Rubin: Seven Chips, Five Rack-Scale Systems, One Vision

Named after the astronomer whose work revealed dark matter, the Vera Rubin platform is Nvidia’s answer to the “we need more than just GPUs” problem. It’s not a single chip — it’s seven chips, five rack-scale systems, and a full supercomputer architecture, all designed to work together as one vertically integrated AI factory.

The standout component: the Vera CPU, Nvidia’s own central processor purpose-built for agentic AI workloads. Huang was characteristically blunt about its potential — “This is already for sure going to be a multi-billion-dollar business for us.”

The real innovation is how Nvidia splits inference into two stages. The Vera Rubin GPU handles “prefill” — transforming your question into AI tokens. Then chips built on technology licensed from Groq for $17 billion handle “decode” — generating the actual response. By separating the workload, each stage gets optimized independently, delivering faster responses at lower cost.

And Nvidia is already looking beyond Vera Rubin. The Feynman architecture, expected in 2028, will include a new CPU called Rosa (after Rosalind Franklin) and next-gen networking chips. The roadmap signals Nvidia’s intent to control every layer of the AI stack — compute, memory, storage, networking, and security.

NemoClaw and the Age of AI Agents

Perhaps the most culturally significant announcement: NemoClaw, Nvidia’s enterprise platform built around the viral open-source project OpenClaw.

Huang called OpenClaw “the most popular open source project in the history of humanity” and declared that “every single company in the world today has to have an OpenClaw strategy.”

That’s Jensen being Jensen, but there’s substance behind it. NemoClaw adds enterprise guardrails that OpenClaw’s consumer-friendly approach lacks — privacy controls, safety mechanisms, and integration with Nvidia’s broader AI infrastructure. It bridges the gap between a fun weekend project and a Fortune 500 deployment.

The message is clear: agentic AI — where systems autonomously execute tasks with minimal human oversight — is no longer a research curiosity. It’s infrastructure.

DLSS 5: AI Rewrites Gaming Graphics

For the gamers (and Nvidia never forgets the gamers), Huang unveiled DLSS 5 — a generational leap using 3D-guided neural rendering to generate photorealistic lighting and materials in real time. Unlike previous versions that primarily upscaled lower-resolution images, DLSS 5 creates visual fidelity that’s indistinguishable from pre-rendered cinematics.

The implications stretch far beyond gaming — architectural visualization, film production, virtual reality. Nvidia described it as “bridging the gap between rendering and reality.” Based on what they showed, that’s not marketing fluff.

Data Centers in Space (Yes, Really)

In what might be the most Jensen Huang announcement ever, Nvidia revealed plans for Space-1 Vera Rubin — AI data centers designed to operate in orbit.

Before you dismiss this as sci-fi theatrics, consider the logic. Earth-bound data centers face mounting constraints: power availability, cooling costs, land use conflicts, and growing regulatory scrutiny. Space offers essentially unlimited solar power and natural vacuum cooling. The Vera Rubin architecture is specifically being designed with orbital thermal and power constraints in mind.

Is this happening tomorrow? No. Is it the kind of moonshot that keeps Nvidia’s narrative compelling? Absolutely.

What This Means for Everyone Else

Business leaders: The inference pivot means AI deployment costs should drop over the next 18–24 months. Vera Rubin’s efficiency gains will filter down to cheaper API calls and more accessible AI services.

Developers: NemoClaw and the OpenClaw ecosystem mean building production-grade AI agents is about to get dramatically easier. The tools are being commoditized, and barriers are dropping fast.

Workers: Huang’s vision of 75,000 Nvidia employees working alongside 7.5 million AI agents by 2036 — a 100:1 ratio — is telling. Whether that’s exciting or terrifying probably depends on your LinkedIn profile.

Investors: The $1 trillion revenue forecast through 2027 is staggering and, for a $5 trillion company, almost expected. The real question is whether Nvidia can maintain dominance as inference opens the door to more competition.

The Bottom Line

GTC 2026 wasn’t about any single product. It was about Nvidia positioning itself as the indispensable infrastructure layer for the age of AI — from gaming PCs to orbital data centers, from training massive models to running billions of AI agent interactions per second.

The shift from training to inference is the most important strategic transition in AI right now, and Nvidia just showed it plans to dominate both sides. With Vera Rubin, NemoClaw, and the Feynman roadmap, the company is building what amounts to a full-stack monopoly on AI computing.

Whether that’s good for innovation long-term is a question worth asking. But for now, the leather jacket keeps winning.